Type-Driven Syntax and Semantics for Composing Meaning Vectors
Stephen Charles Clark · Oxford University Press eBooks · 2013
This chapter presents a theoretical framework that provides an answer to the question of how distributional representations of word meanings should be combined compositionally to produce distributional representations of larger phrases. More specifically, it describes the framework of Coecke et al. (2010) in a form designed to be accessible to computational linguists not familiar with the mathematics of category theory on which the framework is based. It also briefly introduces pregroup grammar, which is the syntactic framework underlying the compositional process. One question the framework does not answer is what the semantic type of a sentence should be; in other words, what vector space sentences should live in. The chapter presents two possible sentence spaces, the first for expository reasons and the second as an example of a sentence space that has been implemented and evaluated in a word-sense disambiguation task.